memoire-m2

My M2 Memoire on mapping class groups & their representations

representations.tex (26577B)

  1 \chapter{Low-Dimensional Representations}\label{ch:representations}
  2 
  3 Having built a solid understanding of the combinatorics of Dehn twists, we are
  4 now ready to attack the problem of classifying the representations of
  5 \(\Mod(\Sigma_g)\) of sufficiently small dimension. As promised, our strategy
  6 is to make use of the \emph{geometrically-motivated} relations derived in
  7 Chapter~\ref{ch:dehn-twists} and Chapter~\ref{ch:relations}.
  8 
  9 Historically, these relations have been exploited by Funar \cite{funar},
 10 Franks-Handel \cite{franks-handel} and others to establish the triviality of
 11 low-dimensional representations, culminating in Korkmaz' \cite{korkmaz} recent
 12 classification of representations of dimension \(n \le 2 g\) for \(g \ge 3\).
 13 The goal of this chapter is to provide a concise account of Korkmaz' results.
 14 
 15 \begin{theorem}[Korkmaz]\label{thm:low-dim-reps-are-trivial}
 16   Let \(\Sigma_g^b\) be the compact surface of genus \(g \ge 1\) with \(b\)
 17   boundary components and \(\rho : \Mod(\Sigma_g^b) \to \GL_n(\mathbb{C})\) be
 18   a linear representation with \(n < 2 g\). Then the image of \(\rho\) is
 19   Abelian. In particular, if \(g \ge 3\) then \(\rho\) is trivial.
 20 \end{theorem}
 21 
 22 Like some of the results we have encountered so far, the proof of
 23 Theorem~\ref{thm:low-dim-reps-are-trivial} is elementary in nature: we proceed
 24 by induction on \(g\) and tedious case analysis. We begin by the base case \(g
 25 = 2\).
 26 
 27 \begin{proposition}\label{thm:low-dim-reps-are-trivial-base-case}
 28   Given \(\rho : \Mod(\Sigma_2^b) \to \GL_n(\mathbb{C})\) with \(n \le 3\), the
 29   image of \(\rho\) is Abelian.
 30 \end{proposition}
 31 
 32 \begin{proof}[Sketch of proof]
 33   Given \(\alpha \subset \Sigma_2^b\), let \(L_\alpha = \rho(\tau_\alpha)\) and
 34   denote by \(E_{\alpha = \lambda} = \{ v \in \mathbb{C}^n : L_\alpha v =
 35   \lambda v \}\) its eigenspaces. Let \(\alpha_1, \alpha_2, \beta_1, \beta_2,
 36   \gamma, \eta_1, \ldots, \eta_{b-1} \subset \Sigma_2^b\) be the curves of the
 37   Lickorish generators from Theorem~\ref{thm:lickorish-gens}, as shown in
 38   Figure~\ref{fig:lickorish-gens-genus-2}.
 39   \begin{figure}
 40     \centering
 41     \includegraphics[width=.2\linewidth]{images/lickorish-gens-gen-2.eps}
 42     \caption{The Lickorish generators for $g = 2$.}
 43     \label{fig:lickorish-gens-genus-2}
 44   \end{figure}
 45 
 46   If \(n = 1\) then \(\rho(\Mod(\Sigma_2^b)) \subset \GL_1(\mathbb{C}) =
 47   \mathbb{C}^\times\) is Abelian. Now if \(n = 2\) or \(3\), by
 48   Proposition~\ref{thm:commutator-normal-gen} it suffices to show \(L_{\alpha_1}
 49   = L_{\beta_1}\), so that \(\tau_{\alpha_1} \tau_{\beta_1}^{-1} \in \ker
 50   \rho\) and thus \(\Mod(\Sigma_2^b)' \subset \ker \rho\) -- i.e.
 51   \(\rho(\Mod(\Sigma_2^b))\) is Abelian. Given the braid relation
 52   \begin{equation}\label{eq:braid-rel-induction-basis}
 53     L_{\alpha_1} L_{\beta_1} L_{\alpha_1}
 54     = L_{\beta_1} L_{\alpha_1} L_{\beta_1},
 55   \end{equation}
 56   this amounts to showing \(L_{\alpha_1}\) and \(L_{\beta_1}\) commute.
 57 
 58   To that end, we exhaustively analyze all of the possible Jordan forms
 59   \begin{align*}
 60     \begin{pmatrix}
 61       \lambda & 0 \\
 62       0       & \lambda
 63     \end{pmatrix}
 64     & \quad{\normalfont(1)}
 65     &
 66     \begin{pmatrix}
 67       \lambda & 0 \\
 68       0       & \mu
 69     \end{pmatrix}
 70     & \quad{\normalfont(2)}
 71     &
 72     \begin{pmatrix}
 73       \lambda & 1 \\
 74       0       & \lambda
 75     \end{pmatrix}
 76     & \quad{\normalfont(3)}
 77     \\
 78     \begin{pmatrix}
 79       \lambda & 0       & 0       \\
 80       0       & \lambda & 0       \\
 81       0       & 0       & \lambda
 82     \end{pmatrix}
 83     & \quad{\normalfont(4)}
 84     &
 85     \begin{pmatrix}
 86       \lambda & 0   & 0   \\
 87       0       & \mu & 0   \\
 88       0       & 0   & \nu
 89     \end{pmatrix}
 90     & \quad{\normalfont(5)}
 91     &
 92     \begin{pmatrix}
 93       \lambda & 1       & 0       \\
 94       0       & \lambda & 1       \\
 95       0       & 0       & \lambda
 96     \end{pmatrix}
 97     & \quad{\normalfont(6)}
 98     \\
 99     \begin{pmatrix}
100       \lambda & 0   & 0   \\
101       0       & \mu & 1   \\
102       0       & 0   & \mu
103     \end{pmatrix}
104     & \quad{\normalfont(7)}
105     &
106     \begin{pmatrix}
107       \lambda & 0       & 0       \\
108       0       & \lambda & 1       \\
109       0       & 0       & \lambda
110     \end{pmatrix}
111     & \quad{\normalfont(8)}
112     &
113     \begin{pmatrix}
114       \lambda & 0       & 0   \\
115       0       & \lambda & 0   \\
116       0       & 0       & \mu
117     \end{pmatrix}
118     & \quad{\normalfont(9)}
119   \end{align*}
120   of \(L_{\alpha_2}\) -- where \(\lambda, \mu, \nu \in \mathbb{C}^\times\) are
121   all distinct. By changing basis we may assume without loss of generality that
122   the matrix \(L_{\alpha_2}\) is exactly its Jordan form, so that \(E_{\alpha_2
123   = \lambda} = \bigoplus_{i \le \dim E_{\alpha_2}} \mathbb{C} e_i\).
124 
125   For cases (1) to (6), we use the change of coordinates principle and
126   different relations to show \(L_{\alpha_1}\) and \(L_{\beta_1}\) lie inside
127   some Abelian subgroup of \(\GL_n(\mathbb{C})\).
128 
129   \begin{enumerate}[leftmargin=1.9cm]
130     \item[\bfseries\color{highlight}(1) \& (4)]
131       By the change of coordinates principle, both \(L_{\alpha_1}\) and
132       \(L_{\beta_1}\) are conjugate to \(L_{\alpha_2} = \lambda\). But the
133       only matrix conjugate to \(\lambda\) is \(\lambda\) itself. Hence
134       \(L_{\alpha_1} = L_{\beta_1} = \lambda \in \mathbb{C}^\times\).
135 
136     \item[\bfseries\color{highlight}(2) \& (5)]
137       Since \(\alpha_2\) is disjoint from both \(\alpha_1\) and \(\beta_1\), it
138       follows from the disjointness relations \([\tau_{\alpha_1},
139       \tau_{\alpha_2}] = [\tau_{\beta_1}, \tau_{\alpha_2}] = 1\) that
140       \(L_{\alpha_1}\) and \(L_{\beta_1}\) preserve the eigenspaces of
141       \(L_{\alpha_2}\), which are all \(1\)-dimensional. Hence \(L_{\alpha_1}\)
142       and \(L_{\beta_1}\) lie inside the subgroup of diagonal matrices -- an
143       Abelian subgroup of \(\GL_n(\mathbb{C})\).
144 
145     \item[\bfseries\color{highlight}(3) \& (6)]
146       As before, it follows from the disjointness relations that \(E_{\alpha_2
147       = \lambda} = \ker (L_{\alpha_2} - \lambda)\) and \(\ker (L_{\alpha_2} -
148       \lambda)^2\) are invariant under both \(L_{\alpha_1}\) and
149       \(L_{\beta_1}\). This implies \(L_{\alpha_1}\) and \(L_{\beta_1}\) are
150       upper triangular matrices with \(\lambda\) along their diagonals. Any
151       such pair of matrices satisfying the braid relation
152       (\ref{eq:braid-rel-induction-basis}) commute.
153   \end{enumerate}
154 
155   Similarly, in case (7) we use the braid relation and the disjointness
156   relations to show \(L_{\alpha_1}\) and \(L_{\beta_1}\) commute -- see
157   \cite[Proposition~5.1]{korkmaz} for a full proof. Cases (8) and (9) require
158   some extra thought. We consider the curve \(\beta_2\). In these cases, the
159   eigenspace \(E_{\alpha_2 = \lambda}\) is \(2\)-dimensional. Since
160   \(L_{\alpha_2}\) and \(L_{\beta_2}\) are conjugate, \(E_{\beta_2 = \lambda}\)
161   is also \(2\)-dimensional -- indeed, conjugate operators have the same Jordan
162   form. Now either \(E_{\alpha_2 = \lambda} = E_{\beta_2 = \lambda}\) or
163   \(E_{\alpha_2 = \lambda} \ne E_{\beta_2 = \lambda}\). We begin by the first
164   case.
165 
166   We claim that if \(E_{\alpha_2 = \lambda} = E_{\beta_2 = \lambda}\) then
167   \(E_{\alpha_2 = \lambda}\) is \(\Mod(\Sigma_2^b)\)-invariant. Indeed, by
168   Observation~\ref{ex:change-of-coordinates-crossing} we can always find \(f,
169   g, h_i \in \Mod(\Sigma_2^b)\) with
170   \begin{align*}
171     f \cdot [\alpha_2]      & = [\alpha_1]
172     &
173     g \cdot [\alpha_2]      & = [\beta_1]
174     &
175     h_i \cdot [\alpha_2]    & = [\alpha_2]    \\
176     f \cdot [\beta_2]   & = [\beta_1]
177     &
178     g \cdot [\beta_2]   & = [\gamma]
179     &
180     h_i \cdot [\beta_2] & = [\eta_i].
181   \end{align*}
182   In particular,
183   \begin{align*}
184     f   \tau_{\alpha_2}    f^{-1}   & = \tau_{\alpha_1}
185     &
186     g   \tau_{\alpha_2}    g^{-1}   & = \tau_{\beta_1}
187     &
188     h_i \tau_{\alpha_2}    h_i^{-1} & = \tau_{\alpha_2}     \\
189     f   \tau_{\beta_2} f^{-1}   & = \tau_{\beta_1}
190     &
191     g   \tau_{\beta_2} g^{-1}   & = \tau_{\gamma}
192     &
193     h_i \tau_{\beta_2} h_i^{-1} & = \tau_{\eta_i}.
194   \end{align*}
195   and thus
196   \begin{align*}
197     E_{\alpha_1 = \lambda}
198     = \rho(f) E_{\alpha_2 = \lambda}
199     & = \rho(f) E_{\beta_2 = \lambda}
200     = E_{\beta_1 = \lambda}
201     \\
202     E_{\beta_1 = \lambda}
203     = \rho(g) E_{\alpha_2 = \lambda}
204     & = \rho(g) E_{\beta_2 = \lambda}
205     = E_{\gamma = \lambda}
206     \\
207     E_{\eta_i = \lambda}
208     = \rho(h_i) E_{\alpha_2 = \lambda}
209     & = \rho(h_i) E_{\beta_2 = \lambda}
210     = E_{\beta_2 = \lambda}.
211   \end{align*}
212   In other words, \(E_{\alpha_1 = \lambda} = E_{\alpha_2 = \lambda} =
213   E_{\beta_1 = \lambda} = E_{\beta_2 = \lambda} = E_{\gamma = \lambda} =
214   E_{\eta_1 = \lambda} = \cdots = E_{\eta_{b-1} = \lambda}\) is invariant
215   under the action of all Lickorish generators.
216 
217   Hence \(\rho\) restricts to a subrepresentation \(\bar \rho :
218   \Mod(\Sigma_2^b) \to \GL(E_{\alpha_2 = \lambda}) = \GL_2(\mathbb{C})\) --
219   recall \(E_{\alpha_2 = \lambda} = \mathbb{C} e_1 \oplus \mathbb{C} e_2\). By
220   case (2), \(\bar \rho(f) = 1\) for all \(f \in \Mod(\Sigma_2^b)'\), given
221   that \(\bar \rho(\Mod(\Sigma_2^b))\) is Abelian. Thus
222   \[
223     \rho(\Mod(\Sigma_2^b)') \subset
224     \begin{pmatrix}
225       1 & 0 & * \\
226       0 & 1 & * \\
227       0 & 0 & *
228     \end{pmatrix}
229   \]
230   lies inside the group of upper triangular matrices, a solvable subgroup of
231   \(\GL_3(\mathbb{C})\). Now by Proposition~\ref{thm:commutator-is-perfect} we
232   get \(\rho(\Mod(\Sigma_2^b)') = 1\): any homomorphism from a perfect group to
233   a solvable group is trivial.
234 
235   Finally, if \(E_{\alpha_2 = \lambda} \ne E_{\beta_2 = \lambda}\) and
236   the Jordan form of \(L_{\alpha_2}\) is given by (8) then the disjointness
237   relations \([\tau_{\alpha_2}, \tau_{\alpha_1}] = [\tau_{\alpha_2},
238   \tau_{\beta_1}] = [\tau_{\beta_2}, \tau_{\alpha_1}] = [\tau_{\beta_2},
239   \tau_{\beta_1}] = 1\) implies that \(L_{\alpha_1}\) and \(L_{\beta_1}\)
240   preserve the eigenspaces of both \(L_{\alpha_2}\) and \(L_{\beta_2}\), so
241   \[
242     0
243     \subsetneq E_{\alpha_2 = \lambda} \cap E_{\beta_2 = \lambda}
244     \subsetneq E_{\alpha_2 = \lambda}
245     \subsetneq V
246   \]
247   is a flag of subspaces invariant under \(L_{\alpha_1}\) and \(L_{\beta_1}\).
248   In this case we can find a basis for \(\mathbb{C}^3\) with respect to which
249   the matrices of \(L_{\alpha_1}\) and \(L_{\beta_1}\) are both upper
250   triangular with \(\lambda\) along the diagonal: take \(v_1, v_2, v_3 \in
251   \mathbb{C}^3\) with \(v_1 \in E_{\alpha_2 = \lambda} \cap E_{\beta_2 =
252   \lambda}\) and \(v_2 \in V_{L_{\alpha_2}}\). Any such pair of matrices
253   satisfying the braid relation (\ref{eq:braid-rel-induction-basis}) commute.
254 
255   Similarly, if \(L_{\alpha_2}\) has Jordan form (9) and \(E_{\alpha_2 =
256   \lambda} \ne E_{\beta_2 = \lambda}\) we use
257   (\ref{eq:braid-rel-induction-basis}) to conclude \(L_{\alpha_1}\) and
258   \(L_{\beta_1}\) commute -- again, see \cite[Proposition~5.1]{korkmaz}. We are
259   done.
260 \end{proof}
261 
262 We are now ready to establish the triviality of low-dimensional
263 representations.
264 
265 \begin{proof}[Proof of Theorem~\ref{thm:low-dim-reps-are-trivial}]
266   Let \(g \ge 1\), \(b \ge 0\), and fix \(\rho : \Mod(\Sigma_g^b) \to
267   \GL_n(\mathbb{C})\) with \(n < 2g\). We want to show
268   \(\rho(\Mod(\Sigma_g^b))\) is Abelian. As promised, we proceed by induction
269   on \(g\). The base case \(g = 1\) is again clear from the fact \(n = 1\) and
270   \(\GL_1(\mathbb{C}) = \mathbb{C}^\times\). The case \(g = 2\) was also
271   established in Proposition~\ref{thm:low-dim-reps-are-trivial-base-case}.
272 
273   Now suppose \(g \ge 3\) and every \(m\)-dimensional representation of
274   \(\Sigma_{g - 1}^{b'}\) has Abelian image for \(m < 2(g - 1)\).
275   Let \(\alpha_1, \ldots, \alpha_g, \beta_1, \ldots, \beta_g, \gamma_1, \ldots,
276   \gamma_{g - 1}, \eta_1, \ldots, \eta_{b-1} \subset \Sigma_g^b\) be the curves
277   from the Lickorish generators of \(\Mod(\Sigma_g^b)\), as in
278   Figure~\ref{fig:lickorish-gens}. Once again, let \(L_\alpha =
279   \rho(\tau_\alpha)\) and denote by \(E_{\alpha = \lambda}\) the eigenspace of
280   \(L_\alpha\) associated to \(\lambda \in \mathbb{C}\). Let \(\Sigma \cong
281   \Sigma_{g - 1}^1\) be the closed subsurface highlighted in
282   Figure~\ref{fig:korkmaz-proof-subsurface}.
283 
284   \begin{figure}[ht]
285     \centering
286     \includegraphics[width=.35\linewidth]{images/lickorish-gens-korkmaz-proof.eps}
287     \caption{The subsurface $\Sigma \subset \Sigma_g^b$.}
288     \label{fig:korkmaz-proof-subsurface}
289   \end{figure}
290 
291   We claim that it suffices to find a \(m\)-dimensional
292   \(\Mod(\Sigma)\)-invariant\footnote{Here we view $\Mod(\Sigma)$ as a subgroup
293   of $\Mod(\Sigma_g^b)$ via the inclusion homomorphism $\Mod(\Sigma) \to
294   \Mod(\Sigma_g^b)$ from Example~\ref{ex:inclusion-morphism}, which can be
295   shown to be injective in this particular case.} subspace \(W \subset
296   \mathbb{C}^n\) with \(2 \le m \le n - 2\). Indeed, in this case \(m < 2(g -
297   1)\) and \(\dim \mfrac{\mathbb{C}^n}{W} = n - m < 2(g - 1)\). Thus both
298   representations
299   \begin{align*}
300     \rho_1 : \Mod(\Sigma) & \to \GL(W) \cong \GL_m(\mathbb{C})
301     &
302     \rho_2 : \Mod(\Sigma) & \to \GL(\mfrac{\mathbb{C}^n}{W})
303     \cong \GL_{n - m}(\mathbb{C})
304   \end{align*}
305   fall into the induction hypothesis -- i.e. \(\rho_i(\Mod(\Sigma))\) is
306   Abelian. In particular, \(\rho_i(\Mod(\Sigma)') = 1\) and we can find some
307   basis for \(\mathbb{C}^n\) with respect to which
308   \[
309     \rho(f) =
310     \left(
311     \begin{array}{c|c}
312       1_m & *         \\ \hline
313         0 & 1_{n - m}
314     \end{array}
315     \right)
316   \]
317   for all \(f \in \Mod(\Sigma)'\) -- where \(1_k\) denotes the \(k \times k\)
318   identity matrix. Since the group of upper triangular matrices is solvable, it
319   follows from Proposition~\ref{thm:commutator-is-perfect} that \(\rho\)
320   annihilates all of \(\Mod(\Sigma)'\) and, in particular, \(\tau_{\alpha_1}
321   \tau_{\beta_1}^{-1} \in \ker \rho\). Now recall from
322   Proposition~\ref{thm:commutator-normal-gen} that \(\Mod(\Sigma_g^b)'\) is
323   normally generated by \(\tau_{\alpha_1} \tau_{\beta_1}^{-1}\), from which we
324   conclude \(\rho(\Mod(\Sigma_g^b)') = 1\), as desired.
325 
326   As before, we exhaustively analyze all possible Jordan forms of
327   \(L_{\alpha_g}\). First, consider the case where we can find eigenvalues
328   \(\lambda_1, \ldots, \lambda_k\) of \(L_{\alpha_g}\) such that the sum \(W =
329   \bigoplus_i E_{\alpha_g = \lambda_i}\) of the corresponding eigenspaces has
330   dimension \(m\) with \(2 \le m \le n - 2\). In this case, it suffices to
331   observe that since \(\alpha_g\) lies outside of \(\Sigma\), each
332   \(E_{\alpha_g = \lambda_i}\) is \(\Mod(\Sigma)\)-invariant: the Lickorish
333   generators \(\tau_{\alpha_1}, \ldots, \tau_{\alpha_{g - 1}}, \tau_{\beta_1},
334   \ldots, \tau_{\beta_{g - 1}}\), \(\tau_{\gamma_1}, \ldots, \tau_{\gamma_{g -
335   2}}\) of \(\Sigma \cong \Sigma_{g - 1}^1\) all commute with
336   \(\tau_{\alpha_g}\) and thus preserve the eigenspaces of its action on
337   \(\mathbb{C}^n\).
338 
339   If no sum of the form \(\bigoplus_i E_{\alpha_g = \lambda_i}\) has dimension
340   lying between \(2\) and \(n - 2\), then there must be at most \(2\) distinct
341   eigenvalues \(\lambda\) of \(L_{\alpha_g}\), and \(\dim E_{\alpha_g =
342   \lambda} = 1, n - 1, n\) for all such \(\lambda\). Hence the Jordan form of
343   \(L_{\alpha_g}\) has to be one of
344   \begin{align*}
345     \begin{pmatrix}
346       \lambda & 0       & 0      & \cdots & 0       & 0       \\
347       0       & \lambda & 0      & \ldots & 0       & 0       \\
348       \vdots  & \vdots  & \vdots & \ddots & \vdots  & \vdots  \\
349       0       & 0       & 0      & \cdots & \lambda & 0       \\
350       0       & 0       & 0      & \cdots & 0       & \lambda
351     \end{pmatrix}
352     & \quad{\normalfont(1)}
353     &
354     \begin{pmatrix}
355       \lambda & 1       & 0      & \cdots & 0       & 0       \\
356       0       & \lambda & 1      & \ldots & 0       & 0       \\
357       \vdots  & \vdots  & \vdots & \ddots & \vdots  & \vdots  \\
358       0       & 0       & 0      & \cdots & \lambda & 1       \\
359       0       & 0       & 0      & \cdots & 0       & \lambda
360     \end{pmatrix}
361     & \quad{\normalfont(2)}
362     \\
363     \begin{pmatrix}
364       \lambda & 0       & 0      & \cdots & 0       & 0       \\
365       0       & \lambda & 0      & \ldots & 0       & 0       \\
366       \vdots  & \vdots  & \vdots & \ddots & \vdots  & \vdots  \\
367       0       & 0       & 0      & \cdots & \lambda & 1       \\
368       0       & 0       & 0      & \cdots & 0       & \lambda
369     \end{pmatrix}
370     & \quad{\normalfont(3)}
371     &
372     \begin{pmatrix}
373       \lambda & 0       & 0      & \cdots & 0       & 0       \\
374       0       & \lambda & 0      & \ldots & 0       & 0       \\
375       \vdots  & \vdots  & \vdots & \ddots & \vdots  & \vdots  \\
376       0       & 0       & 0      & \cdots & \lambda & 0       \\
377       0       & 0       & 0      & \cdots & 0       & \mu
378     \end{pmatrix}
379     & \quad{\normalfont(4)}
380   \end{align*}
381   for \(\lambda \ne \mu\). We analyze the first two sporadic cases
382   individually.
383 
384   \begin{enumerate}
385     \item[\bfseries\color{highlight}(1)]
386       Here we use the change of coordinates principle: each \(L_{\alpha_i},
387       L_{\beta_i}, L_{\gamma_i},  L_{\eta_i}\) is conjugate to \(L_{\alpha_g} =
388       \lambda\), so all Lickorish generators of \(\Mod(\Sigma_g^b)\) act on
389       \(\mathbb{C}^n\) as scalar multiplication by \(\lambda\) as well. Hence
390       \(\rho(\Mod(\Sigma_g^b)) = \langle \lambda \rangle\) is Abelian.
391 
392     \item[\bfseries\color{highlight}(2)]
393       In this case, \(W = \ker (L_{\alpha_g} - \lambda)^2\) is a
394       \(2\)-dimensional \(\Mod(\Sigma)\)-invariant subspace.
395   \end{enumerate}
396 
397   For cases (3) and (4), we consider two situations: \(E_{\alpha_g = \lambda}
398   \ne E_{\beta_g = \lambda}\) or \(E_{\alpha_g = \lambda} = E_{\beta_g =
399   \lambda}\). If \(E_{\alpha_2 = \lambda} \ne E_{\beta_2 = \lambda}\), then \(W
400   = E_{\alpha_g = \lambda} \cap E_{\beta_g = \lambda}\) is a \((n -
401   2)\)-dimensional \(\Mod(\Sigma)\)-invariant subspace: since \(\beta_g\) lies
402   outside of \(\Sigma\) and \(L_{\alpha_g}, L_{\beta_g}\) are conjugate, both
403   \(E_{\alpha_g = \lambda}\) and \(E_{\beta_g = \lambda}\) are
404   \(\Mod(\Sigma)\)-invariant \((n - 1)\)-dimensional subspaces.
405 
406   Finally, we consider the case where \(E_{\alpha_g = \lambda} = E_{\beta_g
407   = \lambda}\). In this situation, as in the proof of
408   Proposition~\ref{thm:low-dim-reps-are-trivial-base-case}, it follows from
409   Observation~\ref{ex:change-of-coordinates-crossing} that there are \(f_i,
410   g_i, h_i \in \Mod(\Sigma_g^b)\) with
411   \begin{align*}
412     f_i \tau_{\alpha_g}    f_i^{-1} & = \tau_{\alpha_i}
413     &
414     g_i \tau_{\alpha_g}    g_i^{-1} & = \tau_{\beta_i}
415     &
416     h_i \tau_{\alpha_g}    h_i^{-1} & = \tau_{\alpha_g}
417     \\
418     f_i \tau_{\beta_g} f_i^{-1} & = \tau_{\beta_i}
419     &
420     g_i \tau_{\beta_g} g_i^{-1} & = \tau_{\gamma_i}
421     &
422     h_i \tau_{\beta_g} h_i^{-1} & = \tau_{\eta_i}
423   \end{align*}
424   and thus
425   \(E_{\alpha_1 = \lambda} = \cdots = E_{\alpha_g = \lambda} = E_{\beta_1 =
426   \lambda} = \cdots = E_{\beta_g = \lambda} = E_{\gamma_1 = \lambda} = \cdots =
427   E_{\gamma_{g - 1} = \lambda} = E_{\eta_1 = \lambda} = \cdots = E_{\eta_{b-1}
428   = \lambda}\).
429 
430   In particular, we can find a basis for \(\mathbb{C}^n\) with respect to
431   which the matrix of any Lickorish generator has the form
432   \[
433     \begin{pmatrix}
434       \lambda & 0       & \cdots & 0       & *      \\
435       0       & \lambda & \ldots & 0       & *      \\
436       \vdots  & \vdots  & \ddots & \vdots  & \vdots \\
437       0       & 0       & \cdots & \lambda & *      \\
438       0       & 0       & \cdots & 0       & *
439     \end{pmatrix}.
440   \]
441   Since the group of upper triangular matrices is solvable and
442   \(\Mod(\Sigma_g^b)\) is perfect, it follows that \(\rho(\Mod(\Sigma_g^b))\) is
443   trivial. This concludes the proof \(\rho(\Mod(\Sigma_g^b))\) is Abelian.
444 
445   To see that \(\rho(\Mod(\Sigma_g^b)) = 1\) for \(g \ge 3\) we note that,
446   since \(\rho\) has Abelian image and thus factors though the Abelianization
447   map \(\Mod(\Sigma_g^b) \to \Mod(\Sigma_g^b)^\ab =
448   \mfrac{\Mod(\Sigma_g^b)}{[\Mod(\Sigma_g^b), \Mod(\Sigma_g^b)]}\). Now recall
449   from Proposition~\ref{thm:trivial-abelianization} that \(\Mod(\Sigma_g^b)^\ab
450   = 0\) for \(g \ge 3\). We are done.
451 \end{proof}
452 
453 Having established the triviality of the low-dimensional representations \(\rho
454 : \Mod(\Sigma_g^b) \to \GL_n(\mathbb{C})\), all that remains for us is to
455 understand the \(2g\)-dimensional representations of \(\Mod(\Sigma_g^b)\). We
456 certainly know a nontrivial example of such, namely the symplectic
457 representation \(\psi : \Mod(\Sigma_g) \to \operatorname{Sp}_{2g}(\mathbb{Z})\)
458 from Example~\ref{ex:symplectic-rep}. Surprisingly, this turns out to be
459 \emph{essentially} the only example of a nontrivial \(2g\)-dimensional
460 representation in the compact case. More precisely,
461 
462 \begin{theorem}[Korkmaz]\label{thm:reps-of-dim-2g-are-symplectic}
463   Let \(g \ge 3\) and \(\rho : \Mod(\Sigma_g^b) \to \GL_{2g}(\mathbb{C})\).
464   Then \(\rho\) is either trivial or conjugate to the symplectic
465   representation\footnote{Here the map $\Mod(\Sigma_g^b) \to
466   \operatorname{Sp}_{2g}(\mathbb{Z})$ is given by the composition of the
467   inclusion morphism $\Mod(\Sigma_g^b) \to \Mod(\Sigma_g)$ with the usual
468   symplectic representation $\psi : \Mod(\Sigma_g) \to
469   \operatorname{Sp}_{2g}(\mathbb{Z})$.} \(\Mod(\Sigma_g^b) \to
470   \operatorname{Sp}_{2g}(\mathbb{Z})\) of \(\Mod(\Sigma_g^b)\).
471 \end{theorem}
472 
473 Unfortunately, the limited scope of these master's thesis does not allow us to
474 dive into the proof of Theorem~\ref{thm:reps-of-dim-2g-are-symplectic}. The
475 heart of this proof lies in a result about representations of the product
476 \(B_3^n = B_3 \times \cdots \times B_3\), which Korkmaz refers to as \emph{the
477 main lemma}.
478 
479 \begin{lemma}[Korkmaz' main lemma]\label{thm:main-lemma}
480   Given \(i = 1, \ldots, n\), denote by 
481   \begin{align*}
482     a_i & = (1, \ldots, 1, \sigma_1, 1, \ldots, 1) &
483     b_i & = (1, \ldots, 1, \sigma_2, 1, \ldots, 1)
484   \end{align*}
485   the \(n\)-tuples in \(B_3^n\) whose \(i\)-th coordinates are \(\sigma_1\) and
486   \(\sigma_2\), respectively, and with all other coordinates equal to \(1\).
487   Let \(m \ge 2n\) and \(\rho : B_3^n \to \GL_m(\mathbb{C})\) be a
488   representation satisfying:
489   \begin{enumerate}
490     \item The only eigenvalue of \(\rho(a_i)\) is \(1\) and its eigenspace is
491       \((m - 1)\)-dimensional.
492     \item The eigenspaces of \(\rho(a_i)\) and \(\rho(b_i)\) associated to the
493       eigenvalue \(1\) do not coincide.
494   \end{enumerate}
495   Then \(\rho\) is conjugate to the representation
496   \begin{align*}
497     B_3^n & \to \GL_m(\mathbb{C}) \\
498     a_i
499     & \mapsto
500     \left(
501     \begin{array}{c|c|c}
502       1_{2(i-1)} & 0                                            & 0 \\ \hline
503                0 & \begin{array}{cc} 1 & 1 \\ 0 & 1 \end{array} & 0 \\ \hline
504                0 & 0                                            & 1_{m-2i}
505     \end{array}
506     \right) \\
507     b_i
508     & \mapsto
509     \left(
510     \begin{array}{c|c|c}
511       1_{2(i-1)} & 0                                             & 0 \\ \hline
512                0 & \begin{array}{cc} 1 & 0 \\ -1 & 1 \end{array} & 0 \\ \hline
513                  0 & 0                                             & 1_{m-2i}
514     \end{array}
515     \right),
516   \end{align*}
517   where \(1_k\) denotes the \(k \times k\) identity matrix.
518 \end{lemma}
519 
520 This is proved in \cite[Lemma 7.6]{korkmaz} using the braid relations. Notice
521 that for \(n = g\) and \(m = 2g\) the matrices in Lemma~\ref{thm:main-lemma}
522 coincide with the action of the Lickorish generators \(\tau_{\alpha_1}, \ldots,
523 \tau_{\alpha_g}, \tau_{\beta_1}, \ldots, \tau_{\beta_g} \in \Mod(\Sigma_g^b)\) on
524 \(H_1(\Sigma_g, \mathbb{C}) \cong \mathbb{C}^{2g}\) -- represented in the standard
525 basis \([\alpha_1], \ldots, [\alpha_g], [\beta_1], \ldots, [\beta_g]\) for
526 \(H_1(\Sigma_g, \mathbb{C})\).
527 \begin{align*}
528   (\tau_{\alpha_i})_* & =
529   \left(
530   \begin{array}{c|c|c}
531     1 & 0                                            & 0 \\ \hline
532     0 & \begin{array}{cc} 1 & 1 \\ 0 & 1 \end{array} & 0 \\ \hline
533     0 & 0                                            & 1
534   \end{array}
535   \right) &
536   (\tau_{\beta_i})_* & =
537   \left(
538   \begin{array}{c|c|c}
539     1 & 0                                             & 0 \\ \hline
540     0 & \begin{array}{cc} 1 & 0 \\ -1 & 1 \end{array} & 0 \\ \hline
541     0 & 0                                             & 1
542   \end{array}
543   \right)
544 \end{align*}
545 
546 Hence by embedding \(B_3^g\) in \(\Mod(\Sigma_g^b)\) via
547 \begin{align*}
548   B_3^g & \to \Mod(\Sigma_g^b)         \\
549   a_i   & \mapsto \tau_{\alpha_i}    \\
550   b_i   & \mapsto \tau_{\beta_i}
551 \end{align*}
552 we can see that any \(\rho : \Mod(\Sigma_g^b) \to \GL_{2g}(\mathbb{C})\) in a
553 certain class of representation satisfying some technical conditions must be
554 conjugate to the symplectic representation \(\Mod(\Sigma_g^b) \to
555 \operatorname{Sp}_{2g}(\mathbb{Z})\) when restricted to \(B_3^g\).
556 
557 Korkmaz then goes on to show that such technical conditions are met for any
558 nontrivial \(\rho : \Mod(\Sigma_g^b) \to \GL_{2g}(\mathbb{C})\). Furthermore,
559 Korkmaz also argues that we can find a basis for \(\mathbb{C}^{2g}\) with
560 respect to which the matrices of \(\rho(\tau_{\gamma_1}), \ldots,
561 \rho(\tau_{\gamma_{g - 1}}), \rho(\tau_{\eta_1}), \ldots,
562 \rho(\tau_{\eta_{b-1}})\) also agree with the action of \(\Mod(\Sigma_g^b)\) on
563 \(H_1(\Sigma_g, \mathbb{C})\), concluding the classification of
564 \(2g\)-dimensional representations.
565 
566 Recently, Kasahara \cite{kasahara} also classified the \((2g+1)\)-dimensional
567 representations of \(\Mod(\Sigma_g^b)\) for \(g \ge 7\) in terms of certain
568 twisted \(1\)-cohomology groups. On the other hand, the representations of
569 dimension \(n > 2g + 1\) are still poorly understood, and fundamental questions
570 remain unsweared. In the short and mid-terms, the works of Korkmaz and Kasahara
571 lead to many follow-up questions. For example,
572 \begin{enumerate}
573   \item In the \(g \ge 3\) case, Korkmaz \cite[Theorem~3]{korkmaz} established
574     the lower bound of \(3g - 3\) for the dimension of an injective linear
575     representation of \(\Mod(\Sigma_g)\) -- if one such representation exists.
576     Can we improve this lower bound?
577 
578   \item What is the minimal dimension for a representation of
579     \(\Mod(\Sigma_g)\) which does not annihilate the entire kernel of the
580     symplectic representation \(\psi : \Mod(\Sigma_g) \to
581     \operatorname{Sp}_{2g}(\mathbb{Z})\)? In particular, do the \((2g +
582     1)\)-dimensional representations classified by Kasahara \cite{kasahara}
583     annihilate all of \(\ker \psi\)?
584 \end{enumerate}
585 
586 These are some of the questions which I plan to work on during my upcoming PhD.